Run AI agents and your own applications side by side in a local workspace
You can now arrange scattered chat logs, documents, browsers, and AI agents on a single screen to work together. HolaOS is an open-source workspace that lets companies or individuals freely combine agents and tools to fit their workflow. All data is stored on the local machine and does not go to external servers, and it is ready to use with a few clicks without complex configuration.
A screen where agents and apps sit side by side
Most AI tools show results only as text within a chat window, but HolaOS displays actual application interfaces next to the agent. Apps like Notion or a browser open on the same screen as the agent, allowing you to watch the agent manipulate the app in real time. If you operate the app directly, that action is immediately passed to the agent's context, so you do not need to explain the situation again every time. You can install apps from the marketplace within the workspace, or create custom apps by specifying a URL and MCP server directly.
Connecting to team chat tools
Most work context lives in chat tools like Slack, Feishu, and DingTalk rather than in documents. HolaOS connects to the chat tools your team already uses, allowing the agent to read actual conversation content directly. You decide which tool and scope to grant access to, so the agent cannot see information outside the approved scope. This ensures that decisions made in meetings or discussions are not missed by the agent and are reflected in the work.
Freely choosing models and agents
You are not tied to a specific model or agent; you can choose tools that fit the nature of the task. You can run Claude Code, Codex, or the built-in HolaOS agent side by side in the same workspace, and they share the same memory, tools, and apps. It provides the latest models such as Kimi K3, GLM 5.2, GPT 5.6, and Claude Opus 5 by default, and you can also bring your own API key or specify a particular endpoint. You can manage all models with a single account or choose a method to control costs with your own keys.
Local-first data control
User control is essential for agents to access tools and files. HolaOS stores workspace files, memory, embeddings, and session history all on the user's local disk. Since no copy of the data exists on a server, the provider cannot see your files or conversations. The agent requires user approval every time it connects a tool or modifies a file, and memory is maintained as plain files that users can read and edit directly. In enterprise environments, it supports SSO, role-based permissions, and audit logs, and can run on-premises or on your own cloud.
Installation and license notes
On macOS, Linux, and WSL environments, you can start installation with a one-line script, and manual installation requires Node.js 24.14.1 and npm. The repository uses a modified Apache 2.0 license, which includes additional conditions for commercial distribution and branding, so you must check the LICENSE file before use. Security issues should be reported privately to the designated email address, and you can track the latest changes in the open-source release notes.
